​
1. AI Events (trackAi)

AI Events are a user input and assistant output pair.

Each AI Event can include:

  • User ID: Identifies the individual user to help you segment analytics.

  • Event Name: Describes what interaction took place (e.g., prompt_submitted, chat_started, summary_generated).

  • Input: The input data or prompt from the user.

  • Output: The AI-generated output or response.

  • Model (optional): Identifies which AI model was used (e.g., GPT-4, Claude, etc.).

  • Convo ID (optional): Allows grouping of related messages into one user conversation.

  • Attachments (optional): Provides additional context for AI interactions, such as images, documents, code snippets, or embedded content. Attachments can be either from the user (e.g. a file or picture) or the assistant (useful if the assistant outputs multiple files).

  • Properties: Flexible tags or metadata (e.g., prompt version, experiment ID, browser type, etc.) that your team can use to filter, segment, and analyze events.

Here’s a visual example of an AI Event:

dawn.track_ai({
  userId: "user123",
  event: "chat_completion",
  model: "gpt-4-turbo",
  input: "What is the weather today in San Francisco?",
  output: "Today's weather in San Francisco is clear and sunny.",
  convoId: "conversation_89",
  attachments: [
    {
      type: "image",
      value: "https://example.com/weather.png",
      role: "output"
    }
  ],
  properties: {
    prompt_version: "v2.3.0",
  }
});

​
2. Signals (trackSignal)

Signals represent explicit or implicit feedback from users about AI outputs. These help your team understand not just that a user saw and engaged with an output, but the quality and usefulness of that output.

If you imagine a coding app like Bolt or v0, signals might include:

  • 👍 or 👎 feedback, along with comments (“this answer was incorrect”)

  • User edits to the generated code (you’ll be able to see what the user edited)

  • Build or syntax errors from the generated code

  • Whether the user tried to deploy the output

Dawn analyzes signals to find patterns and detect hidden issues.

​
3. Normal Analytics Events (track)

In addition to AI-specific observability, Dawn supports traditional analytics tracking. You can track user behaviors such as clicks, page views, conversions, or feature usage. This allows you to seamlessly blend AI interaction analytics with broader product usage metrics.

Example:

dawn.track({
  userId: "user123",
  event: "feature_viewed",
  properties: {
    feature_name: "advanced_search",
    subscription_plan: "premium"
  }
});

​
4. User Identification (setUserDetails)

Associating user metadata allows segmentation, filtering, and deeper understanding of who is interacting with your application:

dawn.setUserDetails({
  userId: "user123",
  traits: {
    email: "user@example.com",
    name: "Jane Doe",
    plan: "enterprise"
  }
});

​
5. Privacy & Security

Dawn takes privacy very seriously. You can choose to enable automated client-side redaction of personally identifiable information (PII):

const dawn = new Analytics({ 
  writeKey: DAWN_API_KEY, 
  redactPii: true 
});

By enabling PII redaction, you ensure sensitive details will never leave your customers’ devices.

​
Coming Soon: AI Tracing 🔥

AI tracing allows you to track detailed AI pipeline execution, capturing step-by-step information of complex multi-model interactions or chained prompts:

  • Easily visualize full trace execution, latency, and errors.

  • Debug and optimize complex prompt chains.

  • Understand intermediate steps that led to a specific generated output.

Was this page helpful?